[PATCH v3] powerpc/64: Option to use ELF V2 ABI for big-endian kernels

[PATCH v3] powerpc/64: Option to use ELF V2 ABI for big-endian kernels

[Nicholas Piggin]

Provide an option to build big-endian kernels using the ELF V2 ABI. This works
on GCC and clang (since about 2014). it is is not officially supported by the
GNU toolchain, but it can give big-endian kernels  some useful advantages of
the V2 ABI (e.g., less stack usage).

Reviewed-by: Segher Boessenkool <segher@kernel.crashing.org>
Signed-off-by: Nicholas Piggin <npiggin@gmail.com>
---
Since v1:
- Improved the override flavour name suggested by Segher.
- Improved changelog wording.

Since v2:
- Improved changelog, help text, to use the name ELF V2 ABI in the spec,
  and clarify things a bit more, suggested by Segher.
- For option name, match the ELF_ABI_v1/2 which is already in the kernel.
- Prefix options with PPC64_ to avoid arch clashes or confusion.
- "elfv2" is the toolchain name of the ABI, so I kept that in the crypto
  perl scripts.

 arch/powerpc/Kconfig            | 21 +++++++++++++++++++++
 arch/powerpc/Makefile           | 15 ++++++++++-----
 arch/powerpc/boot/Makefile      |  4 ++++
 drivers/crypto/vmx/Makefile     |  8 ++++++--
 drivers/crypto/vmx/ppc-xlate.pl | 10 ++++++----
 5 files changed, 47 insertions(+), 11 deletions(-)

diff --git a/arch/powerpc/Kconfig b/arch/powerpc/Kconfig
index 924c541a9260..444867e07039 100644
--- a/arch/powerpc/Kconfig
+++ b/arch/powerpc/Kconfig
@@ -147,6 +147,7 @@ config PPC
 	select ARCH_WEAK_RELEASE_ACQUIRE
 	select BINFMT_ELF
 	select BUILDTIME_TABLE_SORT
+	select PPC64_BUILD_ELF_ABI_V2		if PPC64 && CPU_LITTLE_ENDIAN
 	select CLONE_BACKWARDS
 	select DCACHE_WORD_ACCESS		if PPC64 && CPU_LITTLE_ENDIAN
 	select DYNAMIC_FTRACE			if FUNCTION_TRACER
@@ -541,6 +542,26 @@ config KEXEC_FILE
 config ARCH_HAS_KEXEC_PURGATORY
 	def_bool KEXEC_FILE
 
+config PPC64_BUILD_ELF_ABI_V2
+	bool
+
+config PPC64_BUILD_BIG_ENDIAN_ELF_ABI_V2
+	bool "Build big-endian kernel using ELF V2 ABI (EXPERIMENTAL)"
+	depends on PPC64 && CPU_BIG_ENDIAN && EXPERT
+	default n
+	select PPC64_BUILD_ELF_ABI_V2
+	help
+	  This builds the kernel image using the "Power Architecture 64-Bit ELF
+	  V2 ABI Specification", which has a reduced stack overhead and faster
+	  function calls. This internal kernel ABI option does not affect
+          userspace compatibility.
+
+	  The V2 ABI is standard for 64-bit little-endian, but for big-endian
+	  it is less well tested by kernel and toolchain. However some distros
+	  build userspace this way, and it can produce a functioning kernel.
+
+	  This requires gcc 4.9 or newer and binutils 2.24 or newer.
+
 config RELOCATABLE
 	bool "Build a relocatable kernel"
 	depends on PPC64 || (FLATMEM && (44x || FSL_BOOKE))
diff --git a/arch/powerpc/Makefile b/arch/powerpc/Makefile
index f310c32e88a4..baf477d100b2 100644
--- a/arch/powerpc/Makefile
+++ b/arch/powerpc/Makefile
@@ -92,10 +92,14 @@ endif
 
 ifdef CONFIG_PPC64
 ifndef CONFIG_CC_IS_CLANG
-cflags-$(CONFIG_CPU_BIG_ENDIAN)		+= $(call cc-option,-mabi=elfv1)
-cflags-$(CONFIG_CPU_BIG_ENDIAN)		+= $(call cc-option,-mcall-aixdesc)
-aflags-$(CONFIG_CPU_BIG_ENDIAN)		+= $(call cc-option,-mabi=elfv1)
-aflags-$(CONFIG_CPU_LITTLE_ENDIAN)	+= -mabi=elfv2
+ifdef CONFIG_PPC64_BUILD_ELF_ABI_V2
+cflags-y				+= $(call cc-option,-mabi=elfv2,$(call cc-option,-mcall-aixdesc))
+aflags-y				+= $(call cc-option,-mabi=elfv2)
+else
+cflags-y				+= $(call cc-option,-mabi=elfv1)
+cflags-y				+= $(call cc-option,-mcall-aixdesc)
+aflags-y				+= $(call cc-option,-mabi=elfv1)
+endif
 endif
 endif
 
@@ -144,7 +148,7 @@ endif
 
 CFLAGS-$(CONFIG_PPC64)	:= $(call cc-option,-mtraceback=no)
 ifndef CONFIG_CC_IS_CLANG
-ifdef CONFIG_CPU_LITTLE_ENDIAN
+ifdef CONFIG_PPC64_BUILD_ELF_ABI_V2
 CFLAGS-$(CONFIG_PPC64)	+= $(call cc-option,-mabi=elfv2,$(call cc-option,-mcall-aixdesc))
 AFLAGS-$(CONFIG_PPC64)	+= $(call cc-option,-mabi=elfv2)
 else
@@ -153,6 +157,7 @@ CFLAGS-$(CONFIG_PPC64)	+= $(call cc-option,-mcall-aixdesc)
 AFLAGS-$(CONFIG_PPC64)	+= $(call cc-option,-mabi=elfv1)
 endif
 endif
+
 CFLAGS-$(CONFIG_PPC64)	+= $(call cc-option,-mcmodel=medium,$(call cc-option,-mminimal-toc))
 CFLAGS-$(CONFIG_PPC64)	+= $(call cc-option,-mno-pointers-to-nested-functions)
 
diff --git a/arch/powerpc/boot/Makefile b/arch/powerpc/boot/Makefile
index c53a1b8bba8b..d27e85413c84 100644
--- a/arch/powerpc/boot/Makefile
+++ b/arch/powerpc/boot/Makefile
@@ -41,6 +41,10 @@ endif
 
 BOOTCFLAGS	+= -isystem $(shell $(BOOTCC) -print-file-name=include)
 
+ifdef CONFIG_PPC64_BUILD_ELF_ABI_V2
+BOOTCFLAGS	+= $(call cc-option,-mabi=elfv2)
+endif
+
 ifdef CONFIG_CPU_BIG_ENDIAN
 BOOTCFLAGS	+= -mbig-endian
 else
diff --git a/drivers/crypto/vmx/Makefile b/drivers/crypto/vmx/Makefile
index 709670d2b553..751ffca694aa 100644
--- a/drivers/crypto/vmx/Makefile
+++ b/drivers/crypto/vmx/Makefile
@@ -5,18 +5,22 @@ vmx-crypto-objs := vmx.o aesp8-ppc.o ghashp8-ppc.o aes.o aes_cbc.o aes_ctr.o aes
 ifeq ($(CONFIG_CPU_LITTLE_ENDIAN),y)
 override flavour := linux-ppc64le
 else
+ifdef CONFIG_PPC64_BUILD_ELF_ABI_V2
+override flavour := linux-ppc64-elfv2
+else
 override flavour := linux-ppc64
 endif
+endif
 
 quiet_cmd_perl = PERL $@
       cmd_perl = $(PERL) $(<) $(flavour) > $(@)
 
 targets += aesp8-ppc.S ghashp8-ppc.S
 
-$(obj)/aesp8-ppc.S: $(src)/aesp8-ppc.pl FORCE
+$(obj)/aesp8-ppc.S: $(src)/aesp8-ppc.pl $(src)/ppc-xlate.pl FORCE
 	$(call if_changed,perl)
   
-$(obj)/ghashp8-ppc.S: $(src)/ghashp8-ppc.pl FORCE
+$(obj)/ghashp8-ppc.S: $(src)/ghashp8-ppc.pl $(src)/ppc-xlate.pl FORCE
 	$(call if_changed,perl)
 
 clean-files := aesp8-ppc.S ghashp8-ppc.S
diff --git a/drivers/crypto/vmx/ppc-xlate.pl b/drivers/crypto/vmx/ppc-xlate.pl
index 36db2ef09e5b..b583898c11ae 100644
--- a/drivers/crypto/vmx/ppc-xlate.pl
+++ b/drivers/crypto/vmx/ppc-xlate.pl
@@ -9,6 +9,8 @@ open STDOUT,">$output" || die "can't open $output: $!";
 
 my %GLOBALS;
 my $dotinlocallabels=($flavour=~/linux/)?1:0;
+my $elfv2abi=(($flavour =~ /linux-ppc64le/) or ($flavour =~ /linux-ppc64-elfv2/))?1:0;
+my $dotfunctions=($elfv2abi=~1)?0:1;
 
 ################################################################
 # directives which need special treatment on different platforms
@@ -40,7 +42,7 @@ my $globl = sub {
 };
 my $text = sub {
     my $ret = ($flavour =~ /aix/) ? ".csect\t.text[PR],7" : ".text";
-    $ret = ".abiversion	2\n".$ret	if ($flavour =~ /linux.*64le/);
+    $ret = ".abiversion	2\n".$ret	if ($elfv2abi);
     $ret;
 };
 my $machine = sub {
@@ -56,8 +58,8 @@ my $size = sub {
     if ($flavour =~ /linux/)
     {	shift;
 	my $name = shift; $name =~ s|^[\.\_]||;
-	my $ret  = ".size	$name,.-".($flavour=~/64$/?".":"").$name;
-	$ret .= "\n.size	.$name,.-.$name" if ($flavour=~/64$/);
+	my $ret  = ".size	$name,.-".($dotfunctions?".":"").$name;
+	$ret .= "\n.size	.$name,.-.$name" if ($dotfunctions);
 	$ret;
     }
     else
@@ -142,7 +144,7 @@ my $vmr = sub {
 
 # Some ABIs specify vrsave, special-purpose register #256, as reserved
 # for system use.
-my $no_vrsave = ($flavour =~ /linux-ppc64le/);
+my $no_vrsave = ($elfv2abi);
 my $mtspr = sub {
     my ($f,$idx,$ra) = @_;
     if ($idx == 256 && $no_vrsave) {
-- 
2.23.0

Re: [PATCH v3] powerpc/64: Option to use ELF V2 ABI for big-endian kernels

[Michael Ellerman]

Nicholas Piggin <npiggin@gmail.com> writes:
> Provide an option to build big-endian kernels using the ELF V2 ABI. This works
> on GCC and clang (since about 2014). it is is not officially supported by the
> GNU toolchain, but it can give big-endian kernels  some useful advantages of
> the V2 ABI (e.g., less stack usage).
>
> Reviewed-by: Segher Boessenkool <segher@kernel.crashing.org>
> Signed-off-by: Nicholas Piggin <npiggin@gmail.com>
> ---
> Since v1:
> - Improved the override flavour name suggested by Segher.
> - Improved changelog wording.
>
> Since v2:
> - Improved changelog, help text, to use the name ELF V2 ABI in the spec,
>   and clarify things a bit more, suggested by Segher.
> - For option name, match the ELF_ABI_v1/2 which is already in the kernel.
> - Prefix options with PPC64_ to avoid arch clashes or confusion.
> - "elfv2" is the toolchain name of the ABI, so I kept that in the crypto
>   perl scripts.
>
>  arch/powerpc/Kconfig            | 21 +++++++++++++++++++++
>  arch/powerpc/Makefile           | 15 ++++++++++-----
>  arch/powerpc/boot/Makefile      |  4 ++++
>  drivers/crypto/vmx/Makefile     |  8 ++++++--
>  drivers/crypto/vmx/ppc-xlate.pl | 10 ++++++----
>  5 files changed, 47 insertions(+), 11 deletions(-)

This doesn't build with clang:

  /tmp/aesp8-ppc-dad624.s: Assembler messages:
  /tmp/aesp8-ppc-dad624.s: Error: .size expression for aes_p8_set_encrypt_key does not evaluate to a constant
  /tmp/aesp8-ppc-dad624.s: Error: .size expression for aes_p8_set_decrypt_key does not evaluate to a constant
  /tmp/aesp8-ppc-dad624.s: Error: .size expression for aes_p8_encrypt does not evaluate to a constant
  /tmp/aesp8-ppc-dad624.s: Error: .size expression for aes_p8_decrypt does not evaluate to a constant
  /tmp/aesp8-ppc-dad624.s: Error: .size expression for aes_p8_cbc_encrypt does not evaluate to a constant
  /tmp/aesp8-ppc-dad624.s: Error: .size expression for aes_p8_ctr32_encrypt_blocks does not evaluate to a constant
  /tmp/aesp8-ppc-dad624.s: Error: .size expression for aes_p8_xts_encrypt does not evaluate to a constant
  /tmp/aesp8-ppc-dad624.s: Error: .size expression for aes_p8_xts_decrypt does not evaluate to a constant
  clang: error: assembler command failed with exit code 1 (use -v to see invocation)
  make[4]: *** [/linux/scripts/Makefile.build:349: drivers/crypto/vmx/aesp8-ppc.o] Error 1
  make[4]: *** Waiting for unfinished jobs....
  /tmp/ghashp8-ppc-01aa43.s: Assembler messages:
  /tmp/ghashp8-ppc-01aa43.s: Error: .size expression for gcm_init_p8 does not evaluate to a constant
  /tmp/ghashp8-ppc-01aa43.s: Error: .size expression for gcm_gmult_p8 does not evaluate to a constant
  /tmp/ghashp8-ppc-01aa43.s: Error: .size expression for gcm_ghash_p8 does not evaluate to a constant
  clang: error: assembler command failed with exit code 1 (use -v to see invocation)
  make[4]: *** [/linux/scripts/Makefile.build:349: drivers/crypto/vmx/ghashp8-ppc.o] Error 1
  make[3]: *** [/linux/scripts/Makefile.build:488: drivers/crypto/vmx] Error 2
  make[3]: *** Waiting for unfinished jobs....
  make[2]: *** [/linux/scripts/Makefile.build:488: drivers/crypto] Error 2
  make[2]: *** Waiting for unfinished jobs....
  /linux/drivers/scsi/qla2xxx/qla_nx2.c:3226:1: warning: stack frame size of 2208 bytes in function 'qla8044_collect_md_data' [-Wframe-larger-than=]
  qla8044_collect_md_data(struct scsi_qla_host *vha)
  ^
  1 warning generated.
  make[1]: *** [/linux/Makefile:1722: drivers] Error 2
  make: *** [Makefile:180: sub-make] Error 2


I'm not sure if it can work with clang at all, so I'm happy if the
option depends on !CC_IS_CLANG.

cheers

Re: [PATCH v3] powerpc/64: Option to use ELF V2 ABI for big-endian kernels

[Segher Boessenkool]

Hi!

On Mon, May 18, 2020 at 04:35:22PM +1000, Michael Ellerman wrote:
> Nicholas Piggin <npiggin@gmail.com> writes:
> > Provide an option to build big-endian kernels using the ELF V2 ABI. This works
> > on GCC and clang (since about 2014). it is is not officially supported by the
> > GNU toolchain, but it can give big-endian kernels  some useful advantages of
> > the V2 ABI (e.g., less stack usage).

> This doesn't build with clang:
> 
>   /tmp/aesp8-ppc-dad624.s: Assembler messages:
>   /tmp/aesp8-ppc-dad624.s: Error: .size expression for aes_p8_set_encrypt_key does not evaluate to a constant

What does this assembler code that clang doesn't swallow look like?  Is
that valid code?  Etc.


Segher

Re: [PATCH v3] powerpc/64: Option to use ELF V2 ABI for big-endian kernels

[Nicholas Piggin]

Excerpts from Segher Boessenkool's message of May 18, 2020 10:19 pm:
> Hi!
> 
> On Mon, May 18, 2020 at 04:35:22PM +1000, Michael Ellerman wrote:
>> Nicholas Piggin <npiggin@gmail.com> writes:
>> > Provide an option to build big-endian kernels using the ELF V2 ABI. This works
>> > on GCC and clang (since about 2014). it is is not officially supported by the
>> > GNU toolchain, but it can give big-endian kernels  some useful advantages of
>> > the V2 ABI (e.g., less stack usage).
> 
>> This doesn't build with clang:
>> 
>>   /tmp/aesp8-ppc-dad624.s: Assembler messages:
>>   /tmp/aesp8-ppc-dad624.s: Error: .size expression for aes_p8_set_encrypt_key does not evaluate to a constant
> 
> What does this assembler code that clang doesn't swallow look like?  Is
> that valid code?  Etc.

The .size directive calculation is .text - .opd because the preprocessor 
isn't passing -mabi=elfv2 which makes our _GLOBAL function entry macro 
use the v1 convention. I guess I got the 64-bit vdso wrong as well, it 
should remain in ELFv1.

Thanks,
Nick